Thanks GREQ,
It's understandable but one of the earphones that i tested should exhibit a very flat FR whereas it follows this weird sort of FR algorithm as the other two earphones.
To make things further more complicated l. I ran a free air test of the mic in a reasonably silent environment. The results were unusual. (See the attached image)
I was expecting more flat results
This graph looks like pink noise or a log sweep, which would give that type of diving FR graph. It's not the problem you're asking about, but it won't hurt to use a proper test signal.
As for the 8kHz thing of the previous graph, well the most reasonable way to look at it is to accept that high frequencies will be unreliable. This should be pretty obvious when looking up online graphs, but the reality is even worse because most people sharing their graphs online have first tried to copy some reference(usually another guy making online graphs who copied another guy who ...). So chances are that a lot of those online graphs are compensated in some ways to look less different than the RAW measurement actually are. You just have to pick your poison: 1. Leave the high freqs alone and only use FR graphs from your cheap rig for what they do right, comparisons between 2 IEMs you have measured yourself. AKA reasonable objective use of the data you have.
2. Also try to copy someone's response with a calibration curve, and if it even works in a consistent way(probably won't), make your target known so people have a chance to avoid misinterpreting stuff.
Just know that within what you can find online, some will like you have some recess somewhere near 8kHz, while some actually do all they can to place a resonance around there, so their graphs will show more 8kHz(or near that) than there would be in an expensive dummy head or probably in your own ears. Personally, on my measurement toy rig that I use and trust the most, I have a calibration file I made based mostly my personal hearing and a few other human hearing expectations based on research. And I stopped that calibration file at 8kHz ironically. Above, I've found things to vary too much depending on rather small variables(insertion, tips, coupler). Those differences only get bigger when compared to subjective impressions(2 people may judge an IEM in mostly similar ways in the low end and midrange, but they may describe up to 20dB differences between each other in the upper frequencies. So I've got up to where I thought I was still getting fairly consistent results, and above 8kHz on that rig, "que sera sera".